影片中人物分群方法之研究

The Investigation of Clustering Algorithms for Clustering People in Video

魏良佑(Liang-You Wei)

指導教授 ： 王才沛

國立交通大學/資訊學院/多媒體工程研究所/碩士(2011年)

https://doi.org/10.6842/NCTU.2011.00934

摘要

本文主要是研究影片中人物分群的方法，人物分群最明顯的特徵就是臉部影像，但是影像的解析度、光影、拍攝角度、膚色、場景、分鏡 ( shot ) 卻會大大影響分群的結果，因此本文還利用身體影像以及影片的時間資訊作為分群輔助，我們把相似的人物先合併成演員串列，如此可避免過多雜亂的臉部影像降低效能，並計算串列間的人物相似度矩陣。
　　上述是利用一些客觀的條件處理人物分群，本文還結合叢集整合的概念，試著將這些演員串列分群，可得到較細膩的叢集整合相似度矩陣，藉此了解串列間的相似程度，人物相似度矩陣與叢集整合相似度矩陣兩者依據串列的時間差距決定權重值，分別乘上權重值加總過後，即是最終用來分群的相似度矩陣，搭配階層式凝聚法，我們可以得到最終的分群結果。

關鍵字

人物分群

並列摘要

We investigated clustering algorithm for clustering people in video in this paper. Face image is the most obvious feature of people, but its resolution, luminance, shadow, shooting angle, skin color, and shot, will greatly affect clustering, so we also used the body image and movie time information as the auxiliary. We aggregated the similar people image to the same actor sequence, it can avoid that many disorderly face images reduced performance, and then we computed person similarity matrix between sequences for use.
　Above-mentioned use some objective condition to cluster people, and we also integrated the concept of cluster ensemble, and we tried to cluster the actor sequences. The ensemble similarity matrix is more exquisite than the person similarity matrix. It can help us to realize the similarity between actor sequences. Person similarity matrix and ensemble similarity matrix product their own weight which be computed according to the difference of time, and we sum up the two products of similarity matrix and their own weight, taking it as the final similarity matrix for clustering. We used the final similarity matrix on hierarchical agglomeration to find the final clustering.

並列關鍵字

people clustering

參考文獻

[1] W. Y. Zhao, R. Chellappa, P. J. Phillips, and A. Rosenfeld,“Face recognition: a literature survey,”ACM Computing Surveys (CSUR), vol. 35, no. 4, pp. 399-458, 2003.

[2] J. Tao and Y. P. Tan,“Efficient clustering of face sequences with application to character-based movie browsing,”Proc. IEEE International Conference on Image, pp.1708~1711, 2008.

[4] E. El-Khoury, C. Senac, and P. Joly, "Face-and-Clothing Based People Clustering in Video Content," Proc. International Multimedia Conference on Multimedia Information Retrieval, pp. 295-304, 2010.

[5] P. Huang, Y. Wang, and M. Shao, "A New Method for Multi-view Face Clustering in Video Sequence," Proc. IEEE International Conference on Data Mining Workshops, pp. 869-873, 2008.

[6] K. Yamamoto, O. Yamaguchi, and H. Aoki,“Fast face clustering based on shot similarity for browsing video,”Progress in Informatics, pp. 53-62, 2010.